`<locale>`: Repair `std::collate<unsigned short>` #5361

muellerj2 · 2025-03-25T18:48:16Z

This repairs collate<unsigned short> by adding two missing unsigned short specializations for _LStrcoll and _LStrxfrm when we're building the STL. Fixes #5236.

There are similar specializations for a few functions in <xlocale>. I guess _LStrcoll and _LStrxfrm were missed because they are located in a different header.

Since collate is the only user of these functions, I also moved them from <xlocinfo> to <locale> just above collate.

As for why this also fixes collate<wchar_t> = collate<unsigned short> under non-native wchar_t when linking to the DLL, this is because the the locale facets are constructed in the DLL here:

STL/stl/src/wlocale.cpp

Lines 58 to 88 in f2a2933

    
           // moved from locale to ease subsetting 
        
           using _Tu1  = ctype<unsigned short>; 
        
           using _Tu2  = num_get<unsigned short>; 
        
           using _Tu3  = num_put<unsigned short>; 
        
           using _Tu4  = numpunct<unsigned short>; 
        
           using _Tu5  = collate<unsigned short>; 
        
           using _Tu6  = messages<unsigned short>; 
        
           using _Tu7  = money_get<unsigned short>; 
        
           using _Tu8  = money_put<unsigned short>; 
        
           using _Tu9  = moneypunct<unsigned short, false>; 
        
           using _Tu10 = moneypunct<unsigned short, true>; 
        
           using _Tu11 = time_get<unsigned short>; 
        
           using _Tu12 = time_put<unsigned short>; 
        
           using _Tu13 = codecvt<unsigned short, char, _Mbstatet>; 
        
           void __CLRCALL_OR_CDECL locale::_Locimp::_Makeushloc(const _Locinfo& lobj, locale::category cat, _Locimp* ptrimp, 
        
               const locale* ptrloc) { // setup wide part of a new locale 
        
               ADDFAC(_Tu1, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu2, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu3, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu4, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu5, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu6, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu7, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu8, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu9, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu10, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu11, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu12, cat, ptrimp, ptrloc); 
        
               ADDFAC(_Tu13, cat, ptrimp, ptrloc); 
        
           }

Thus, collate's virtual functions originate from the DLL as well. These virtual functions call _LStrcoll and _LStrxfrm, but the specializations for unsigned short were missing since the DLL is built with native wchar_t.

In the new test, we have to skip collate::transform() tests when there is IDL mismatch between TU and linked DLL.

stl/inc/locale

tests/std/tests/VSO_0000000_instantiate_iterators_misc/test.compile.pass.cpp

tests/std/tests/GH_005236_collate_facet/test.cpp

StephanTLavavej · 2025-04-16T05:15:31Z

Thanks! I updated the product code's preprocessor logic, pushed significant changes to the test, and updated the PR description accordingly - please meow if I messed something up.

The general principle that we now follow (although our legacy codebase didn't) is that while /Zc:wchar_t- should work (fake wchar_t), normal users of /Zc:wchar_t (real wchar_t) should not experience unsigned short as anything special. Only the STL's separately compiled code has to worry about real and fake wchar_t simultaneously.

StephanTLavavej · 2025-04-22T10:13:18Z

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

StephanTLavavej · 2025-04-22T21:40:35Z

Thanks for fixing the world's best library with the world's worst compiler option! 😹 💚 🎉

<locale>: Repair std::collate<unsigned short>

5723b00

muellerj2 requested a review from a team as a code owner March 25, 2025 18:48

github-project-automation bot added this to STL Code Reviews Mar 25, 2025

github-project-automation bot moved this to Initial Review in STL Code Reviews Mar 25, 2025

StephanTLavavej added the bug Something isn't working label Mar 25, 2025

StephanTLavavej self-assigned this Mar 25, 2025

StephanTLavavej added 6 commits April 15, 2025 15:09

Merge branch 'main' into locale-repair-collate-ushort

cb62695

Guard the specializations with #ifdef _CRTBLD.

0561095

Adjust comment to "unsigned short sequences".

92455d5

Mention <locale> with angle brackets in test comment.

31ae1f6

Remove squirrels, test wchar_t as an ordinary user (real or fake).

c7fd832

Add const.

64dfb6a

StephanTLavavej reviewed Apr 16, 2025

View reviewed changes

StephanTLavavej approved these changes Apr 16, 2025

View reviewed changes

StephanTLavavej removed their assignment Apr 16, 2025

StephanTLavavej moved this from Initial Review to Ready To Merge in STL Code Reviews Apr 16, 2025

StephanTLavavej moved this from Ready To Merge to Merging in STL Code Reviews Apr 22, 2025

StephanTLavavej merged commit 42cee97 into microsoft:main Apr 22, 2025
39 checks passed

github-project-automation bot moved this from Merging to Done in STL Code Reviews Apr 22, 2025

muellerj2 deleted the locale-repair-collate-ushort branch May 31, 2025 21:49

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

`<locale>`: Repair `std::collate<unsigned short>` #5361

`<locale>`: Repair `std::collate<unsigned short>` #5361

Uh oh!

muellerj2 commented Mar 25, 2025 •

edited by StephanTLavavej

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

StephanTLavavej commented Apr 16, 2025

Uh oh!

StephanTLavavej commented Apr 22, 2025

Uh oh!

Uh oh!

StephanTLavavej commented Apr 22, 2025

Uh oh!

Uh oh!

	// moved from locale to ease subsetting
	using _Tu1 = ctype<unsigned short>;
	using _Tu2 = num_get<unsigned short>;
	using _Tu3 = num_put<unsigned short>;
	using _Tu4 = numpunct<unsigned short>;
	using _Tu5 = collate<unsigned short>;
	using _Tu6 = messages<unsigned short>;
	using _Tu7 = money_get<unsigned short>;
	using _Tu8 = money_put<unsigned short>;
	using _Tu9 = moneypunct<unsigned short, false>;
	using _Tu10 = moneypunct<unsigned short, true>;
	using _Tu11 = time_get<unsigned short>;
	using _Tu12 = time_put<unsigned short>;
	using _Tu13 = codecvt<unsigned short, char, _Mbstatet>;

	void __CLRCALL_OR_CDECL locale::_Locimp::_Makeushloc(const _Locinfo& lobj, locale::category cat, _Locimp* ptrimp,
	const locale* ptrloc) { // setup wide part of a new locale
	ADDFAC(_Tu1, cat, ptrimp, ptrloc);
	ADDFAC(_Tu2, cat, ptrimp, ptrloc);
	ADDFAC(_Tu3, cat, ptrimp, ptrloc);
	ADDFAC(_Tu4, cat, ptrimp, ptrloc);
	ADDFAC(_Tu5, cat, ptrimp, ptrloc);
	ADDFAC(_Tu6, cat, ptrimp, ptrloc);
	ADDFAC(_Tu7, cat, ptrimp, ptrloc);
	ADDFAC(_Tu8, cat, ptrimp, ptrloc);
	ADDFAC(_Tu9, cat, ptrimp, ptrloc);
	ADDFAC(_Tu10, cat, ptrimp, ptrloc);
	ADDFAC(_Tu11, cat, ptrimp, ptrloc);
	ADDFAC(_Tu12, cat, ptrimp, ptrloc);
	ADDFAC(_Tu13, cat, ptrimp, ptrloc);
	}

<locale>: Repair std::collate<unsigned short> #5361

<locale>: Repair std::collate<unsigned short> #5361

Uh oh!

Conversation

muellerj2 commented Mar 25, 2025 • edited by StephanTLavavej Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

StephanTLavavej commented Apr 16, 2025

Uh oh!

StephanTLavavej commented Apr 22, 2025

Uh oh!

Uh oh!

StephanTLavavej commented Apr 22, 2025

Uh oh!

Uh oh!

`<locale>`: Repair `std::collate<unsigned short>` #5361

`<locale>`: Repair `std::collate<unsigned short>` #5361

muellerj2 commented Mar 25, 2025 •

edited by StephanTLavavej

Loading